Jump to content

Deep Blue (chess computer)

From Alfawiki
Deep Blue
Company typeProject / Supercomputer
IndustryComputer hardware, Artificial intelligence, Chess
PredecessorDeep Thought
Founded1989
FounderFeng-hsiung Hsu, Murray Campbell, Arthur Joseph Hoane, Jerry Brody
Headquarters
Armonk, New York
,
United States
ProductsChess-playing supercomputer
ParentIBM

Deep Blue was a specialized IBM RS/6000 SP supercomputer engineered specifically for playing chess. It holds the distinction of being the first computer to win both a game and an entire match against a reigning world chess champion under official time controls. The project originated in 1985 at Carnegie Mellon University with the development of ChipTest, which was later renamed Deep Thought after transitioning to IBM, before finally becoming Deep Blue in 1989. In a six-game series in 1996, it faced world champion Garry Kasparov, achieving one victory, two draws, and three defeats. Following enhancements in 1997, Deep Blue triumphed over Kasparov by winning two games and drawing three, marking a pivotal achievement in the history of artificial intelligence and inspiring numerous publications and films.

History

During his doctoral studies at Carnegie Mellon University, Feng-hsiung Hsu created a chess-playing supercomputer called ChipTest. This system claimed victory at the North American Computer Chess Championship in 1987, with its successor, Deep Thought, winning the same title in 1988. After completing his PhD in 1989, Hsu teamed up with Murray Campbell at IBM Research to develop a machine capable of defeating a world chess champion. By 1990, programmer Arthur Joseph Hoane and IBM veteran Jerry Brody had joined the project.

Following Deep Thought’s two-game defeat against Kasparov in 1989, IBM organized a naming contest for the machine. The name "Deep Blue," proposed by Peter Fitzhugh Brown, was selected as a tribute to IBM’s nickname "Big Blue." A smaller version of Deep Blue played against Grandmaster Joel Benjamin, who was hired to assist in creating Deep Blue's opening move database for its matches with Kasparov. In 1995, a prototype of Deep Blue participated in the eighth World Computer Chess Championship, drawing with Wchess but losing to Fritz.

One of Deep Blue’s two hardware racks is preserved by the National Museum of American History and was previously displayed as part of their Information Age exhibit. The other rack is exhibited at the Computer History Museum within the "Artificial Intelligence and Robotics" section. The development story of Deep Blue is chronicled in several works, including Behind Deep Blue: Building the Computer that Defeated the World Chess Champion by Feng-hsiung Hsu.

Deep Blue versus Kasparov

After Deep Thought’s loss to Garry Kasparov in 1989, Deep Blue confronted Kasparov on two further occasions. Their initial encounter took place from 10 to 17 February 1996, during which Deep Blue became the first computer to win a game against a reigning world champion under standard time controls. Nevertheless, Kasparov won three games and drew two more out of the remaining five, securing the match with a score of 4–2.

Prior to their rematch in May 1997, Deep Blue underwent significant hardware upgrades that doubled its processing speed. This enhanced version defeated Kasparov with a final score of 3½–2½, clinching victory in the last game after Kasparov lost his positional advantage. This event represented the first occasion on which a computer bested a reigning world champion in an official tournament match. The upgraded system typically analyzed six to eight moves ahead, sometimes extending its search depth to over twenty moves. Experts estimate that each additional half-move of foresight contributes an increase of approximately 50-70 Elo rating points.

During the opening game of their second match, a software glitch caused Deep Blue to enter an unintended loop and exit by selecting a valid move at random. Kasparov misread this anomaly as evidence of superior intelligence from the machine, which negatively affected his performance afterward, although he denied that anxiety influenced his play.

Following his loss, Kasparov alleged that human assistance had influenced Deep Blue during their second game; however, IBM refuted these claims by stating that human involvement occurred solely between games. Kasparov requested another rematch, but IBM declined after dismantling Deep Blue. According to match rules, program adjustments were permitted between games to rectify weaknesses identified during play. Kasparov sought access to the machine’s log files; IBM initially refused but eventually made them publicly available online.

The prize money awarded for the 1997 event amounted to $700,000 for the Deep Blue team and $400,000 for Kasparov. Additionally, Carnegie Mellon University granted $100,000 through an award established by professor Edward Fredkin for the first computer program to defeat a reigning world chess champion.

Chess

Kasparov initially described Deep Blue as an "alien opponent," but later reduced its intelligence to that of "an alarm clock." Two grandmasters who competed against Deep Blue likened it to "a wall advancing towards you." Although Hsu retained the rights to utilize the Deep Blue design independently from IBM, he chose not to accept Kasparov's offer for a rematch. The 2003 documentary Game Over: Kasparov and the Machine examined Kasparov's accusations of IBM cheating, with some interviewees implying that IBM’s investment was partly motivated by a desire to enhance its stock market performance.

Other games

Following Deep Blue’s triumph, AI expert Omar Syed developed Arimaa, a game designed to be straightforward for humans yet difficult for computers. Nevertheless, by 2015, AI had surpassed strong Arimaa players. Since then, artificial intelligence programs have made significant strides in other complex board games with active competitive scenes. A prominent example is the AlphaGo series (AlphaGo, AlphaGo Zero, AlphaZero), which defeated elite Go players between 2016 and 2017, highlighting AI’s expanding commercial and technological influence in strategic gaming.

Computer science

For computer scientists like Deep Blue’s developer Campbell, chess has long served as a crucial benchmark for assessing artificial intelligence capabilities. IBM’s success in defeating a reigning world chess champion marked a milestone in AI development. Moreover, Deep Blue helped popularize the use of games as platforms to demonstrate AI progress, influencing subsequent projects such as IBM Watson and AlphaGo.

Deep Blue was capable of analyzing approximately 200 million positions per second and was the first machine to compete against a world chess champion under official conditions. It functioned as an advanced expert system, relying on rules crafted by both chess masters and computer scientists. In contrast, contemporary engines like Leela Chess Zero employ reinforcement learning to train neural networks that create their own internal strategies instead of relying on human-defined heuristics.

In a 2006 contest between Deep Fritz and world champion Vladimir Kramnik, the software ran on a dual-core Intel Xeon 5160 processor, evaluating around 8 million positions per second and achieving an average search depth of 17 to 18 plies in middlegame situations using heuristic methods; it won the match with a score of 4–2.

Software

Deep Blue operated on the AIX operating system and its chess program was written in C. Its evaluation function began as a broad model with numerous parameters optimized through analysis of thousands of master-level games. This function was subsequently segmented into 8,000 specialized components. The opening repertoire contained more than 4,000 positions and incorporated data from 700,000 grandmaster games, while the endgame database included all possible five-piece or fewer endgames along with many six-piece endgames. An "extended book" compiled entire grandmaster matches, allowing the system to merge its capacity to examine 200 million positions per second with strategic opening move selection.

Prior to the second encounter with Kasparov, grandmaster Joel Benjamin enhanced the program’s rule set. The opening book was curated by grandmasters Miguel Illescas, John Fedorowicz, and Nick de Firmian. When Kasparov requested access to Deep Blue’s previous game records for preparation purposes, IBM refused, leading him to rely on studying popular PC chess software instead.

Hardware

Deep Blue incorporated specialized VLSI chips designed to parallelize the alpha–beta search algorithm, serving as a prime example of symbolic AI. Its formidable capability stemmed largely from its computational power, operating on an IBM RS/6000 SP supercomputer featuring a highly parallel structure with 30 PowerPC 604e processors alongside 480 custom-built 600 nm CMOS VLSI "chess chips." These chips were responsible for running the chess expert system and included FPGAs intended for updates (though these were ultimately not utilized), all contained within two separate cabinets. Each chess chip was composed of four distinct units: a move generator, a smart-move stack, an evaluation function, and search control logic. The move generator itself was an 8x8 combinational logic network designed to emulate the chessboard.

By 1997, Deep Blue had been enhanced to rank as the 259th most powerful supercomputer on the TOP500 list, delivering 11.38 GFLOPS as measured by the LINPACK benchmark. This upgraded model was capable of assessing 200 million positions per second—twice the number processed by its 1996 version.

See also

  • Anti-computer tactics, strategies targeting predictable computer behavior
  • IBM Watson, expert in human language question answering
  • Mechanical Turk, a historical chess-playing deception from the 18th and 19th centuries
  • X3D Fritz, another chess engine that drew against Kasparov
  • Rematch, a 2024 television miniseries depicting the 1997 confrontation

Bibliography


External links

Retrieved from "https://www.alfawiki.org/index.php?title=Deep_Blue_(chess_computer)&oldid=28426"